Drinking water quality and social vulnerability linkages at the system level in the United States

The flow chart describes the main data inputs, data analyses, and outputs of the study (figure 1). We used a newly developed dataset of tiers of CWS service area boundaries developed by EPIC/SimpleLab [36] (supporting information, SI, section 2.8; figure 2). We used CWSs with Tier 1 data which are based on states which provide explicit boundaries (15 states) and Tier 2 data which are produced by a matching algorithm to link CWSs to boundaries of a town or city. We excluded Tier 3 data which are estimated using a statistical model and considered less accurate. The Tier 1 and 2 data totaled ∼28 128 CWSs, representing 58% of all CWSs covering 85% of CWS population served (263 million out of 308 million people) and distributed across ∼3000 out of 3143 counties in the contiguous US (CONUS). We do not have CWS boundaries for all tribal systems.

Zoom In Zoom Out Reset image size

Our analysis focused on HB violations of the EPA SDWA which fall into three categories (table S2)

• (1)  
  Maximum contaminant levels (MCLs),
• (2)  
  Maximum residual disinfectant levels, and
• (3)  
  Treatment technique [37].

In cases where contaminant quantification is technically or economically infeasible, treatment rules, such as the Groundwater Rule, Surface Water Treatment Rules, and Lead & Copper Rule have been established. Additional data on SDWA regulations along with the health effects are provided in SI, section 2.1. The groupings we used resulted in ten specific rule violations, as follows: (1) arsenic, (2) nitrates, (3) inorganics, (4) organics [volatile and semivolatile organic compounds], (5) radionuclides, (6) Revised Total Coliform Rule, (7) DBP Rules, (8) Lead & Copper Rule, (9) Surface Water Treatment Rules, and (10) Ground Water Rule. For some analyses, we aggregated rules: 'any HB violation' refers to all HB violation rules, 'any inorganic' includes arsenic, nitrate, and inorganics rule violations.

Violation data from 2018 to 2020 (12 quarters) were selected because EPA regulations were fairly stable during this time. Analysis focused on systems active on 15 April 2021 (time of Safe Drinking Water Information System [SDWIS] data download). The number of systems with HB violations was considered along with populations served by these CWSs. Populations served by CWSs are defined as very small (⩽500 people), small (501–3300), medium (3301–10 000), large (10 001–100 000), and very large (⩾100 000) (table S3). The majority of CWSs (81%) are very small to small systems. The majority of CWSs are sourced by groundwater (72%), and are generally small, but most of the population (73%) is served by systems sourced by surface water.

Alongside the drinking water quality violation data, we modified the CDC SVI that was developed for natural hazards to focus on social vulnerability parameters that have either been identified in the literature to relate to disparities in drinking water quality or ranked highly in our preliminary random forest classification modeling (section 2.5). Differences between mSVI (mSVI, 15 US Census variables grouped into three themes) and the original CDC SVI are described in SI, section 2.4 (figure 3). These differences include expansion of the Socioeconomic Status theme to include per-capita income in mSVI because most states define DACs using income data [23]. Data on household composition were excluded from mSVI because attributes related to household composition were considered more relevant to hazard response. Minority status was expanded in mSVI to include Hispanic, Black, and Asian because many previous studies found linkages between drinking water quality noncompliance and ethnicity [1, 25, 28]. Most of the attributes related to housing in mSVI were retained from the CDC SVI but transportation 'no vehicle' was excluded because transportation is more important for hazard response. Population density was added to the mSVI as a proxy for rurality because previous studies found rurality to be important for drinking water quality violations [24].

Zoom In Zoom Out Reset image size

Tract-level data were used in this analysis rather than higher resolution block-level data because the margins of error associated with blocks are considered too large. Tract level census variables were area weighted to CWS service areas. Following CDC methodology, we calculated mSVI using a percent rank across all census tracts resulting in normalized mSVI values ranging from 0 to 1. All tribal systems are currently classified as disadvantaged according to the EPA and therefore were assumed to have high levels of social vulnerability for the purposes of this study.

We used balanced random forest (BRF) to assess linkages between social vulnerability parameters described in the previous section and occurrence of a HB SDWA violation, which is a binary response variable (±HB violation) (table S3). Because HB violations represent a highly imbalanced dataset with only ∼10% of all CWSs in violation [16], BRF was selected because it is suitable for dealing with imbalanced datasets [38]. Details of the approach are provided in SI, section 2.5(b). The BRF analysis was based on occurrence of any single SDWA violation during the period 2018–2020, excluding multiple violations of the same rule from the model. Recurrence of violations over time was analyzed separately (section 2.5).

Output from BRF was used with SHapley Additive exPlanations (SHAPs) to assess the relative contribution of explanatory variables (social parameters) on the probability of a CWS having a violation. TreeSHAP, a variant of SHAP for tree-based models, was used to estimate SHAP values for tree models as it can handle dependent features [39, 40]. Feature SHAP values represent the impact of the feature (social parameters) on the probability of a HB violation. Purely random data would have a probability of a violation of 0.5. SHAP values represent the absolute change in probability beyond 0.5 associated with each explanatory variable. Model output data, including the testing datasets, SHAP rankings, and selected partial dependence plots, can be found in tables S9–S19.

We further examined temporal variability in HB violations, focusing on recurrence of violations based on the total number of quarters in violation from 2018 to 2020 (total of 12 quarters). Violations of some SDWA rules did not typically recur, including Surface Water Treatment Rule, Ground Water Rule, Revised Total Coliform Rule, and Lead & Copper Rule (table S21). For recurring violations, we calculated the relationship between recurrence and median mSVI of CWSs in violation.

Approximately one in ten people were exposed to any HB violation (∼26 million out of 260 million people, table 2) over the three-year period from 2018 through 2020, corresponding to 11% of all Tier 1 and Tier 2 CWSs considered (3165/28128). This percentage of CWSs with any HB violation is similar to that for all CWS across the CONUS (10%; ∼5000/∼49000 CWSs).

The dominant cause of violations (by percent of CWSs with HB violations) is DBPs during water treatment, resulting in DBP Rule violations (38%) (figure 4). The next highest category is 'any inorganic + radionuclides' (which we define as combined arsenic, nitrate, and inorganics plus radionuclide rule violations) accounting for 22% of CWS violations, primarily related to naturally occurring (e.g. arsenic, uranium) and anthropogenic (e.g. nitrate) contaminants. Violations of the Groundwater Rule account for 21% of CWS violations, Revised Total Coliform Rule for 12%, Surface Water Treatment Rule for 11%, and Lead and Copper Rule for 7%. Organic rule violations account for the smallest fraction of violations (0.6% of CWSs) which is surprising considering that organic violations include an extensive list of 53 chemicals. Additionally, environmental justice issues of water quality often highlight community proximity to industrial facilities and superfund sites [28] rather than formation of DBPs and naturally occurring inorganics and radionuclides.

Zoom In Zoom Out Reset image size

The mSVI map shows high mSVI in census tracts throughout the southern US and many areas in the western US. Low mSVI is shown in much of the Midwest and Northeast (figure 3(b)). Drinking water quality violations disproportionately impact socially vulnerable populations with the majority (71%) of the population impacted by any HB violation residing in the high mSVI tercile of CWSs included in this study (∼28 000 CWSs), whereas much lower populations fall into the middle (19%) and lowest (10%) terciles of mSVI (table 3). Therefore, ∼7 out of 10 people experiencing any HB violation are classified by high mSVI.

The relationship between population impacted and mSVI tercile varies by rule; 61% of the population impacted by arsenic rule violations plot within the high mSVI tercile, 68% for DBP rules, 49% for nitrate, and 42% for radionuclides (table 3). Violations of the remaining rules (Surface Water Treatment, Groundwater, Revised Total Coliform, Lead & Copper, and organics) correspond to 40%–82% of high mSVI populations.

Tribal CWSs were also disproportionately impacted by HB violations with ∼3 in 10 people impacted by a HB violation relative to 1 in 10 people impacted across the CONUS (figure 5). Tribal systems with violations are primarily located in Arizona, New Mexico, and California. In other words, a total of 18% of tribal CWSs had at least one HB violation, corresponding to 28% of the 1 million people served by tribal systems over 2018–2020. HB violations of tribal systems consisted primarily of the Ground Water Rule (44%), followed by Revised Total Coliform Rule (22%), and arsenic rule (16%) system violations (figure 5). These values may be underestimating HB violations due to deficiencies in SDWA enforcement in tribal systems [41].

Zoom In Zoom Out Reset image size

Our analysis shows that many social parameters are linked to drinking water quality violations, with relative importance of different parameters varying by rule (figure 6). Socioeconomic parameters do not always rank most highly based on SHAP analyses, which contrasts with many state and federal tools relying almost exclusively on MHI to define DACs [23] or emphasizing socioeconomics as primary metrics for environmental justice disparities [22, 32].

Zoom In Zoom Out Reset image size

Considering any HB violation, all three themes (socioeconomic, race + language, and demographics/housing) were represented in the top five social parameters in the SHAP results (figure 6(a)). Any HB violation was positively related to vacant homes (depopulation) increasing the probability of incurring a violation by 6.3%. The direction of the impact (positive or negative) is shown by the partial dependence plots (table S9(d)). Population density reduced the probability of any HB violation, also ranking high, and may reflect linkages between violations and rurality consistent with previous results[24]. Race and language were also important factors with Asian reducing the probability (4.6%) and Hispanic increasing the probability (2.8%) of any HB violation incurred by a CWS. These relationships may reflect Asian populations being more concentrated in urban settings and Hispanic population more broadly distributed throughout the southwest and southcentral US. By contrast, socioeconomic factors generally ranked lower, with per capita income ranking 5th, poverty ranking 8th, and MHI ranking 9th. Violations of arsenic, nitrate, any inorganic, and radionuclide rules exhibited a prominent linkage with race and language particularly with Hispanic population (positive impact, increasing probability of violation by 8.4%–18%) and limited English (positive, ∼7%) within the top five parameters (figures 6(b), (c); tables S10–S13). Population density also ranked highly for these violations, with negative impacts on violation probabilities ranging from 6.7% to 14%. The predominance of these factors may reflect the geographic distribution of these violations, in mostly rural areas within the southwest and southcentral US [16]. Socioeconomic factors rank lower than race and language for these violations.

Socioeconomic factors are more prominently associated with DBP rule violations, with MHI, disabled, and no high school diploma positively related to probability of DBP violations (each ∼6%) (figure 6(d)). Vacant housing was also positively related to probability of DBP violations (8.1%) (table S14). This is consistent with DBP rule violations occurring in low-income, suburban systems [16], relative to the geography of inorganics which predominate in rural systems sourced by groundwater. Surface Water Treatment rule violations were most clearly explained by socioeconomic factors (figure 6(e)). There was little consistency in the ranking of social parameters for the remaining drinking water quality violation rules with all three themes (socioeconomic, race + language, demographic) contributing to violations (figure 6, table S17).

Recurrence of any HB violation was also strongly linked to mSVI, particularly for certain rules (figure 7), indicating that CWSs with the most persistent, systemic violations are also the most socially vulnerable. Median mSVI increased with recurrence of any HB violation based on quarterly data for 2018–2020 (R = 0.73; figure 7(a)). All correlation coefficients in this study are highly significant, with p values <0.0001. Moderate to strong relationships between median mSVI of violating systems and number of quarters in violation were found for inorganic contaminants (arsenic, R = 0.73; nitrate, R = 0.44) and treatment issues (DBP rules, R = 0.54) (figures 7(b)–(d)) despite the vastly different formation and mobilization mechanisms of these contaminants. Analysis of initial and recurring violations of selected rules reveals that systems with lower socioeconomic status and higher proportions of minority groups are more likely to experience initial and recurring drinking water violations [1].

Zoom In Zoom Out Reset image size

In contrast, violations of most of the other rules, including Surface Water Treatment, Ground Water, Lead & Copper, and Revised Total Coliform rules did not extend beyond the first few quarters, suggesting that violations of treatment technique rules may be more readily addressed across all CWSs, regardless of social vulnerability status (table S21).

The new federal drinking water infrastructure funding requires states to allocate ⩾49% of the funding to DACs [23]. States have substantial discretion in defining DACs, resulting in large variations in DAC definitions across the US. Currently, 49 of 51 states incorporate MHI in their definitions with eight states relying solely on MHI or a similar metric. New York and Maryland are the only two states that currently incorporate EJ criteria, including unemployment, population decline, and race [23]. Results of this analysis suggest that a broader definition of DACs, beyond MHI, should be considered as mSVI captures 3× more of the affected population than MHI considering the most vulnerable 25% of mSVI compared to the lowest 25% of MHI (figure 8).

Zoom In Zoom Out Reset image size

Our analysis highlights the broader dimensions of social vulnerability, beyond pure economic metrics, which explain linkages with noncompliant CWSs. To date, both state and federal environmental justice programs employ a single metric, primarily based on economic factors, in defining social vulnerability [22, 32]. Additional social parameters with existing data sources at the local level should be considered, including rurality, race and language, population trends, and housing type and tenure when defining DACs, to better capture disparities in access to safe drinking water quality. In addition, the emphasis on drinking water quality in this study suggests that environmental justice tools should expand their consideration of drinking water quality issues beyond point sources and include naturally occurring geogenic and non-point source anthropogenic contaminants [16, 17], as applied in the California Community Water Center tool [42]. While there are inherent limitations to using social vulnerability indices, policy makers and states currently use less rigorous metrics to define DACs and distribute funds. Accordingly, we propose the potential use of mSVI for drinking water quality violations, which is informed by literature and data analytics, to better capture inequities in SDWA violation incidence and recurrence in the US. The proposed mSVI is currently exploratory and would need to be validated/tested more rigorously to lean into a recommendation on wider usage beyond this study.

Continual expansion of the database of CWS service area boundaries [27] will provide more comprehensive coverage to match to mSVI across the CONUS to better understand state-specific demographics and socioeconomic drivers of SDWA violations. Tribal systems should also be considered in much more detail in efforts to conduct more refined spatial analyses. While we evaluated HB violations of tribal CWSs, many tribal regions lack access to any water infrastructure, highlighting the multiple dimensions of water inequities [3], beyond drinking water quality.

This study has important implications for solving drinking water quality issues across the U.S., especially in DACs. Incidents and recurrence of drinking water quality violations are strongly linked to social vulnerability.

One solution to address water quality violations in socially vulnerable communities is to prevent source water contamination [28]; however, this may have limited relevance to tackling pervasive naturally occurring geogenic contaminants (e.g. arsenic, radionuclides) that impact socially vulnerable communities in the Southwest and Southcentral US [16]. Similarly, nitrate contamination is generally a non-point source contaminant that remains in groundwater for decades and is linked to agriculture in much of the Southwest and Southcentral US [35]. It is difficult for CWSs to address pervasive arsenic [43] and nitrate [35] contamination in groundwater given the lack of surface water availability to blend with groundwater or provide an alternative water source in the overexploited, semiarid Southwest and Southcentral US. The only option is often installation of treatment systems; however, this is extremely challenging for small rural systems serving marginalized communities with limited financial, managerial, and technical capacity. Point-of-use and point-of-entry systems are and must increasingly be considered as an alternative to centralized treatment systems [44].

In addition, potential merger, regionalization or CWS consolidation partners may bias away from systems serving low-income and minority communities due to high transaction costs [45], highlighting the need to ensure that DAC definitions effectively prioritize the most vulnerable communities for funding and assistance efforts. We show that recurrence of violations is a critical issue; however, the new federal infrastructure funding does not allow funding for ongoing operation and maintenance of CWSs, which is paramount to ensuring long-term system compliance. More promisingly, the federal EPA's new Water System Restructuring Assessment Rule will require primacy agencies (states and tribes) to assess the feasibility of restructuring for noncompliant systems and provides incentives for implementation [46].

Ultimately, many of the challenges DACs face in achieving and maintaining compliance result from limited financial, managerial, and technical capacity. Many lack the staff to apply for funding, the expertise to operate infrastructure, and the governance structures to ensure inclusive representation of community needs. Providing more technical assistance and management options to these systems and to community-based organizations working with DACs is essential to enhance these programs and overcome recurrent drinking water violations which continue to impair public health in the US.